The finding

The central finding was not the model. It was input data uncertainty making the NPV of the same project swing by more than R$ 20,000 per hectare. A governance problem, not an algorithm problem.

What was built

An automated extraction workflow, using OCR, text mining, and regular expressions, that structured 13,000 records from 332 heterogeneous documents of the Brazilian forestry sector.

Publication

Final project of the MBA in Data Science, AI and Analytics (USP/Esalq), published as a peer-reviewed article in Revista Estratégias e Soluções, titled “Governança de Dados e Risco Financeiro no Setor Florestal” (Data Governance and Financial Risk in the Forestry Sector). The article compares Brazil and Finland and is co-authored with Juliano Souza Vasconcelos.

DOI: 10.22167/2675-6528-202602216